tpd2 and standardised tobacco packaging what impacts have … · 2020-05-12 · updating health...

- 1 -

TPD2 and standardised

tobacco packaging — What

impacts have they had so far?

November 2017

Europe Economics is registered in England No. 3477100. Registered offices at Chancery House, 53-64 Chancery Lane, London WC2A 1QU.

Whilst every effort has been made to ensure the accuracy of the information/material contained in this report, Europe Economics assumes no

responsibility for and gives no guarantees, undertakings or warranties concerning the accuracy, completeness or up to date nature of the

information/analysis provided in the report and does not accept any liability whatsoever arising from any errors or omissions.

© Europe Economics. All rights reserved. Except for the quotation of short passages for the purpose of criticism or review, no part may be used or

reproduced without permission.

Contents

Summary .................................................................................................................................................................................... 1

1 Introduction .................................................................................................................................................................... 2

2 Data and modelling approach ..................................................................................................................................... 3

2.1 Introduction ........................................................................................................................................................... 3

2.2 Primary data sources ........................................................................................................................................... 3

2.3 Measuring TPD2 and plain-packaging requirements ..................................................................................... 3

2.4 Modelling approach .............................................................................................................................................. 4

3 Impacts on Consumption ............................................................................................................................................ 7

3.1 Introduction ........................................................................................................................................................... 7

3.2 Tobacco consumption in the UK and France ................................................................................................ 7

3.3 Simple trend model .............................................................................................................................................. 8

3.4 More sophisticated tests ................................................................................................................................... 10

4 Time-series smoking prevalence model for the UK ........................................................................................... 14

4.1 Introduction ......................................................................................................................................................... 14

4.2 Smoking prevalence England ............................................................................................................................ 14

4.3 Simple trend model ............................................................................................................................................ 14

4.4 More sophisticated tests ................................................................................................................................... 15

5 Conclusion .................................................................................................................................................................... 17

6 Appendix ....................................................................................................................................................................... 19

6.1 Simple trend models for tobacco consumption (FR and UK) and smoking prevalence (UK). ........ 19

6.2 Pure time-series models for tobacco consumption (FR and UK) and smoking prevalence (UK). .. 23

6.3 Simultaneous equation models of tobacco consumption (FR and UK) and smoking prevalence (UK). 29

Summary

- 1 -

Summary

This report was commissioned from Europe Economics by JTI (Japan Tobacco International). In it, we have

considered the impacts of TPD2 and plain packs requirements, introduced in the UK and France between

2016 and 2017, upon tobacco consumption and prevalence, in three types of models: simple linear trend

models, time series models and simultaneous equations models.

A simple trend model considers whether the prior trend in tobacco consumption or prevalence was

changed at the time TPD2 and plain packs requirements were introduced.

A time series model makes the simple trend model more sophisticated by considering the possibility that

the prior evolution was more complex than simply a trend, possibly reflecting lags, seasonality and moving

averages, and also taking account of prices.

A simultaneous equations model allows for the possibility that TPD2 and plain packs requirements might

have their effects on prevalence or consumption either directly or via having an impact on prices, which

in turn had an impact on consumption or prevalence.

We have found that the data so far published indicate no statistically significant impact in the sorts of models

we have used:

No statistically significant impacts on prevalence in the UK.

No statistically significant impacts on consumption in the UK or in France.

We also note that at the time of the impact assessment accompanying TPD2 it was anticipated that there

would be an impact by this point. In the government’s TPD2 impact assessment of 2015 it provides estimates

of projected prevalence levels and projected reductions associated with TPD2. 1 These were for a reduction

of 1.9 per cent in both smoking prevalence and smoking consumption over 5 years, applied linearly implying

a reduction of 0.38 per cent in the 1st year, equivalent to 0.08 percentage points reduction in prevalence in

the first year (2016) if prevalence would have been 19.7 per cent.

No such impact is yet identifiable in the data.

1 See paragraphs 73-74 of Tobacco Products Directive (TPD) IA No: 3131, 29/06/2015, https://www.bma.org.uk/-

/media/files/pdfs/working%20for%20change/policy%20and%20lobbying/uk%20consultations/po-tobacco-products-

directive-impact-assesment-2015-09-01.pdf

Introduction

- 2 -

1 Introduction

In February 2014 the European Union agreed a revised Tobacco Products Directive2, often referred to as

“TPD2”. As well as introducing various other measures, such as restrictions on the advertising of electronic

cigarettes and other vaping devices, TPD2 introduced a series of additional restrictions on the packaging of

tobacco products, such as:

making 20 the minimum number of cigarettes per cigarette pack, and 30 grams the minimum weight for

roll-your-own tobacco packs;

updating health warnings and requiring that combined (picture and text) health warnings cover 65 per

cent of the front and back of cigarette and roll-your-own tobacco packages; and

banning certain descriptors on packaging of tobacco products (such as “natural” and “organic”).

At around the same time France and the UK have adopted additional measures imposing standardised packing

of tobacco products (“plain packs” which we refer to as “PP” requirements). Such measures involve precise

restrictions with regards to:

the banning of all brand elements with the exception of the name which has to appear in a standardized

font and size;

the material, size, shape and opening mechanism of packaging;

the colour of packaging and cigarettes; and

the font, colour, size, case and alignment of text on packs.

Europe Economics was commissioned by Japan Tobacco International to assess any impacts yet discernible

from TPD2 and plain packs upon tobacco consumption in France and the UK, and smoking prevalence in the

UK.

Our analysis indicates that, based on data up to and including August 2017, the data so far published indicate

no statistically significant impact, in the sorts of models we have used, between TPD2+PP requirements and

tobacco consumption in France or the UK. Similarly, again in the models we have used, those data indicate

no statistically significant relationship between the presence of TPD2+PP requirements and smoking

prevalence in the UK.

This report is structured as follows.

Section 2 describes the data used for the analysis and intuitively explains our modelling approach.

Section 3 sets out the analysis on tobacco consumption in France and the UK.

Section 4 presents the analysis on the prevalence of smoking in the UK.

A technical appendix is included at the end of this report, setting out various mathematical points in more

detail, for reference.

2 Directive 2014/40/EU ec.europa.eu/health/sites/health/files/tobacco/docs/dir_201440_en.pdf

Data and modelling approach

- 3 -

2 Data and modelling approach

2.1 Introduction

In this section we describe the data sources the analysis relies upon and provide an intuitive description of

the modelling approach adopted.

2.2 Primary data sources

The raw data underpinning the analysis has been provided by JTI and consist of monthly retail prices and

numbers of sticks for each tobacco product sold in France and the UK. The volume of sticks data covers the

period Jan 2008 — Aug 2017, whilst price data is available only for the period Jan 2012 — Aug 2017.

The data is provided at a high level of disaggregation — information on the number of sticks sold and retail

prices is provided at the product level (i.e. for each separate sub-brand and package size). In this respect, for

the purpose of the analysis, the data has been aggregated so as to obtain:

The number of sticks sold for the whole tobacco market (consisting of cigarettes, roll-your-own

products, and make-your-own products).3

The average price of tobacco products which has been calculated as the weighted average price across

all products, whereby the weights are proportional to the number of sticks sold. These average prices

have then been expressed in a “20 sticks” equivalent from.

For the UK, our analysis also incorporates information on smoking and vaping prevalence. The underlying

hypothesis is that the emergence of electronic cigarettes may have affected the dynamics of the tobacco

market in the UK and should therefore be incorporated in the analysis of our variables of interest for testing

(namely downtrading and smoking prevalence).4 Data on smoking and vaping prevalence in the UK were

obtained from the Smoking Toolkit Study (STS).5 The STS includes up-to-date data tracking national smoking

and vaping patterns, as well as cessation-related behaviours.

2.3 Measuring TPD2 and plain-packaging requirements

In France and the UK, the TPD2 directive and plain-packaging requirements ( “TPD2+PP”) were adopted in

May 2016. Following that transposition there was a transition period at the end of which all tobacco products

sold needed to be compliant with the new regulation. The deadlines after which all products were obliged to

comply with TPD2+PP requirements were: January 2017 for France; and May 2017 for the UK. This means

that, during the implementation period, TPD2+PP products were sold next to products with the “old”

branded packaging format. Therefore, for the purpose of the analysis, the degree of implementation of

TPD2+PP requirements can be interpreted in terms of the penetration rate of TPD2+PP compliant products

in the market.

3 For RYO/MYO products the raw data provides also a sticks-equivalent conversion. 4 We emphasize that our models do not assume that the growth of vaping has led to a reduction in tobacco prevalence

or consumption. Rather, vaping is included as a potential explanatory variable for the statistics to inform us of its

impact. 5 See http://www.smokinginengland.info/latest-statistics/

http://www.smokinginengland.info/latest-statistics/


- 4 -

For France and the UK we have fairly detailed information on the evolution of the actual penetration rate of

TPD2+PP products within each tobacco brand and, therefore we have accurate information on the

penetration of TPD2-PP products. These are illustrated in the chart below.

Figure 2.1: TPD2/PP Penetration (whole market)

2.4 Modelling approach

The class of models we rely upon in this report are so-called “time series” models, in which we first attempt

to infer the underlying evolution through time of the variables we are interested in (e.g. prevalence,

consumption,) and then consider whether (and if so to what extent) the introduction of TPD2 and plain

packs disturbed that underlying evolution path. Such models answer the question “What would have

happened had TPD2 and/or plain packs not been introduced?” roughly as “The variables we are interested in

would have continued to evolve through time in the ways they had done prior to the introduction of TPD2

and/or plain packs.”

More detail on time series models

The most well-known “time series” relationship is probably a trend. Suppose that in country X the

consumption of cigarettes fell steadily by 0.5 percentage points each year for 20 years before TPD2 and plain

packs were introduced. Then (assuming, naïvely, for the purposes of illustration, that no other factors were

found to be relevant) it might be reasonable to assume that if TPD2 and plain packs had not been introduced,

the consumption of cigarettes in country X would have continued to drop by 0.5 percentage points each

year. If in fact the consumption of cigarettes fell consistently by 0.75 percentage points each year after TPD2

and plain packs were introduced, we might (again assuming naïvely that no other factors were found to be

relevant) infer that the introduction of TPD2 and plain packs had been correlated with an acceleration of

0.25 percentage points each year in the decline in cigarette consumption.

Trends are only one form of time series relationship. Others include lags (the value of a variable in any one

year is some multiple of its value in the year before) and moving averages (which can take a number of forms

— e.g. a moving average, over three periods, of the variation of a variable from its trend value). In a time

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

UK FR


- 5 -

series model we use our data to describe, as closely as possible, the evolution through time of the variable(s)

we are interested in via such time series relationships. The impact of a measure such as TPD2 and plain packs

is, then, the way in which it leads to changes in the evolution of the variable(s) of interest relative to these

time series relationships.

One important point to note here is that in this sort of time series model it is not possible to disentangle the

impacts of two measures introduced at the same time. For example, in France and the UK TPD2 and plain

packs are introduced together (as of May 2016). All that the model can do is to say whether the time series

relationships were disturbed from May 2016 onwards. It cannot say why or what proportion of any impacts

should be attributed to different things that happened at that same date.

A further point to note here is that the use of time series relationships automatically accounts for seasonal

effects in our data (where they exist). For example if the value of a variable in a particular month of the year

is always, say, elevated, then it will have some relationship to its value one year before, which the time series

tests will automatically incorporate.

Pure time series versus simultaneous equations models

Our time series models infer time series relationships whilst controlling for prices. For example, consider a

model such as the impact of TPD2 and plain packs upon consumption of tobacco products. It is natural to

suppose that the average price of tobacco products might affect their consumption. So it is natural that such

relative prices feature in the model.

But now suppose that as well as having a direct impact on tobacco consumption, measures such as TPD2 and

plain packs also affected tobacco prices. Then there would be two routes by which impacts would occur: the

direct one and the indirect one. If the model controls for price changes it will miss part of the impact — the

impact that arises indirectly by causing the prices themselves to change. Such a distortion is illustrated below:

Figure 2.2: Distorted impact found in a model where prices are controlled for but the measure has an

impact on prices themselves

We can correct for this by what are commonly referred to as econometric “simultaneous equations”

techniques. In our simultaneous equations models we estimate the impact of the measure (here, TPD2 and

plain packs) upon prices and consumption (or other relevant variables in other models, such as prevalence)

at the same time as measuring the direct impact of the measure upon consumption (or other variables).

Doing so allows us to capture both impacts:

The direct impact of TPD2+PP on consumption; and

The indirect impact of TPD2+PP on consumption, which is enabled by the direct impact of TPD2+PP on

prices.

These are further illustrated in the figure below:


- 6 -

Figure 2.3: Undistorted impact found in a model where impacts on prices and via prices are estimated

simultaneously with direct impacts

Impacts on Consumption

- 7 -

3 Impacts on Consumption

3.1 Introduction

In this section we first provide an overview of smoking consumption in the UK and France. We then present

a series of statistic models: to make ideas concrete we start by illustrating ideas using simple (indeed, in some

senses naïve) models and then we move to more sophisticated ones.

3.2 Tobacco consumption in the UK and France

The following charts indicates the total number of sticks (these include both cigarette and RYO/MYO

products) sold to consumers in the UK and France.

Figure 3.1: Number of sticks (millions) sold in the UK

3,000

3,500

4,000

4,500

5,000

5,500

2012M

01

2012M

03

2012M

05

2012M

07

2012M

09

2012M

11

2013M

01

2013M

03

2013M

05

2013M

07

2013M

09

2013M

11

2014M

01

2014M

03

2014M

05

2014M

07

2014M

09

2014M

11

2015M

01

2015M

03

2015M

05

2015M

07

2015M

09

2015M

11

2016M

01

2016M

03

2016M

05

2016M

07

2016M

09

2016M

11

2017M

01

2017M

03

2017M

05

2017M

07

Number of sticks Average number of sticks over the previous 12 months


- 8 -

Figure 3.2: Number of sticks (millions) sold in France

We can notice from Figure 2.1 that in the UK there has been a declining trend in the number of RMC sticks

sold since 2012, with perhaps some suggestion of a flattening in the rate of fall since mid-2016. In France the

sales of sticks decreased up to 2014 (albeit on a declining trend less pronounced than in the UK), but

somewhat stabilised afterwards.

3.3 Simple trend model

We begin with a very simple “naïve trend analysis”. In such a test we try to explain the evolution of tobacco

consumption with merely a simple linear trend, and test whether there was a break in the series after the

transposition of the TPD2+PP regulation. More specifically we have tested for the following types of breaks:

A break in absolute consumption levels;

A break in consumption trend;

A break in, both absolute consumption levels and consumption trend.

Such a model would not itself demonstrate whether the results of later, more advanced models were correct

or not, but they would help us to understand to what extent results for later, more sophisticated models

with extra control variables were a matter of those extra control variables

Validating (i.e. producing similar results to);

Reinforcing (i.e. producing results with the same sign but with larger coefficients than);

Removing (i.e. eliminating results that were there before in)

Reversing (i.e. producing results in the opposite direction to); or

Adding to (i.e. producing results that were not initially there for)

the results of simple trend analysis.

For France we ran a model with a single linear trend but in such model the trend was not statistically

significant. This suggests that tobacco consumption in France is better described as fluctuating around a

constant mean rather than following a single linear trend. Furthermore none of the tests described above

indicates the presence of a statistically significant break in a naïve trend analysis of the consumption series

for France.

3,000

3,500

4,000

4,500

5,000

5,500

6,000

6,500

7,000

2012M

01

2012M

03

2012M

05

2012M

07

2012M

09

2012M

11

2013M

01

2013M

03

2013M

05

2013M

07

2013M

09

2013M

11

2014M

01

2014M

03

2014M

05

2014M

07

2014M

09

2014M

11

2015M

01

2015M

03

2015M

05

2015M

07

2015M

09

2015M

11

2016M

01

2016M

03

2016M

05

2016M

07

2016M

09

2016M

11

2017M

01

2017M

03

2017M

05

2017M

07

Number of sticks Average number of sticks over the previous 12 months


- 9 -

In contrast, the analysis for the UK suggests that tobacco consumption can be described — in statistically

significant terms — as following a declining trend. The break tests we conducted suggest that the consumption

series for the UK can be described either as having a break in its level (resulting in a higher level thereafter

— i.e. more tobacco consumption, not less) at around May 2016 or as having a break in trend (resulting in a

lower rate of decrease in tobacco consumption — i.e. more tobacco consumption, not less) after May 2016.

Table 3.1: Results of a naïve trend model with a break in the level of consumption (UK)

Coefficient Std. Error Prob.

Constant 5,600,000,000 141,000,000 0.0000

Trend -19,045,013 1,897,574 0.0000

Dummy (break in levels) 242,000,000 87,804,627 0.0076

Table 3.2: Results of a naïve trend model with a break in consumption trend (UK)


Constant 5,600,000,000 1.41E+08 0.0000

Trend -19,121,538 1,910,859 0.0000

Trend * dummy (break in trend) 2,298,611 829,214 0.0073

We also checked whether there might be a break affecting both consumption levels and consumption trend

simultaneously. We do not find such a break to be statistically significant. So there is either a rise in tobacco

consumption levels, or a fall in the rate of decrease, but not both.

Table 3.3: Results of a naïve trend model with a break in consumption level and trend (UK)

More specifically, the results of Table 3.1 indicate that the introduction of TPD2+PP is associated (in this

simple model) with a permanent increase in consumption of around 240 million sticks. Similarly, the results

of Table 3.2 indicate that, whilst before the introduction of TPD2+PP, the monthly reduction in number of

sticks sold in the UK was, on average, of the order of 19 million (the coefficient of the trend variable in Table

3.2), since May 2016 the average monthly reduction is of the order of approximately 17 million (the coefficient

of the coefficient of the trend variable in Error! Reference source not found. plus that of the trend-

times-dummy variable, i.e. -19.1m+2.3m = -16.8m).

Thus in this simple model, the introduction of TPD2+PP results in higher tobacco consumption.

We shall now explore to what extent this basic result from the data is validated, reinforced, removed,

reversed or added to in more sophisticated models. We shall see that for our preferred class of models the


Constant 5,610,000,000 144,000,000 0.0000

Trend -19,147,176 1,938,408 0.0000

Dummy (break in levels) -140,000,000 1,220,000,000 0.9089

Trend * dummy (break in trend) 3,621,667 11,541,262 0.7547


- 10 -

result is removed (i.e. there is no impact), but in some of our cross-check models for the UK it is validated

(i.e. the result that TPD2+PP is associated with an increase in consumption in the UK is repeated).6

3.4 More sophisticated tests

As noted, the naïve trend model does not take into account certain statistical properties of the data. For

example, a statistical inspection of the data indicates that tobacco consumption has a strong seasonal

component (i.e. tobacco consumption higher in certain months of the year) and the simple trend model does

not account for this. Moreover, the introduction of TPD2+PP regulation is modelled with a crude dummy

variable which does not reflect the fact that the penetration of TPD2+PP compliant products in the market

has increased gradually over the implantation period. Finally, the model does not account for other economic

variables (such as prices or household income) that may also affect consumption.

We have therefore implemented more sophisticated tests to model the evolution of tobacco consumption.

The first class of such tests includes pure-time series models, the second class includes simultaneous equation

models tests. These are presented in turns below.

3.4.1 Time-series models

The time-series models we use aim to explain monthly percentage changes in the number of sticks sold by

time series components, monthly percentage changes in prices and the penetration of TPD2+PP products in the

market. The models also include monthly dummy variables to account for seasonal patterns in the data.

If we were conducting tests over a large number of years, as well as adjusting for prices it would be important

to adjust, also, for household income. Our data is, however, relatively high frequency (monthly) but over only

a few years (five). That means both that adjusting for household income is less crucial (it is less likely to be

statistically significant over a short time period), less available (GDP data is available quarterly, not monthly,

not available for the last two data points in our series and GDP per capita data is less reliable than GDP data

at very short time periods because population estimates tend not to be reliable at very high frequency), and

less straightforwardly interpreted (it is not clear that monthly fluctuations in GDP per capita would, even in

theory, be expected to drive fluctuations in consumption insofar as such fluctuations reflected annual income

stream volatility (e.g. self-employment revenue streams, bonuses, etc). Nonetheless, we have cross-checked

the results that follow using models that allowed for the presence of GDP per capita, also.7 None of the

results below changed materially.

In all the models we have tested there is a strong seasonal pattern to consumption. More formally, monthly

dummies are strongly significant and account for a sizable portion of the variation in the data, or, in other

words, information on the specific calendar month is a very strong predictor of the tobacco consumption

taking place in that month.

Finally, in addition to seasonal patterns, changes in prices and, changes in the penetration of TPD2+PP

compliant products, there might be other factors that are important in explaining the evolution of

consumption behaviour. When data evolve through time, it is common to model them using a class of what

are referred to as “autoregressive–integrated–moving–average” (ARIMA) process. Such a process attempts

to describe the behaviour of variables by exploiting any systematic relationship between a variable’s current

value and its past values.

6 Specifically, we find an increase of around 100m in the level of number of sticks sold in the UK. That compares with

the increase of 200m we found in Table 3.1. We regard these are broadly similar impacts, in this context. 7 GDP per capita is statistically significant in the preferred consumption model and prevalence model for the UK, but

leaves the results intact. In the preferred consumption model for France GDP per capita is not statistically significant.


- 11 -

Two key components in ARIMA processes are the “autoregressive” (AR) term and the moving average (MA)

term.8 The “autoregressive” term describes how the present value of the variable depends on its previous

value at some point in the past (say the previous month, or three months ago, or twelve months ago). The

moving average term describes how the noisy fluctuations around the current values of the variable depend

on the noisy fluctuations observed in the past.9 ARIMA models are particularly useful because they can

provide an accurate description of a time-series variable by using only the information contained in the

variables itself, i.e. without the need for additional control variables. However, in our setting the purpose of

including an ARIMA process is that of capturing residual patterns in consumption data that cannot be

explained by the other explanatory variables included in the model (namely seasonal dummies, prices and

TPD-2 penetration). Among the available range of ARIMA models, the “best” ARIMA model (i.e. the one

with the correct orders for the AR and MA terms) can be selected algorithmically based on standard statistical

tests.10

We present below the results for the time series model that our algorithm indicated should be our preferred

time series model for both the UK and France (for presentational purposes the results for the monthly

seasonal dummies are omitted from the tables).

Table 3.4: Pure time-series model of tobacco consumption for the UK

Coeff. Std. Error P-value

% Change in the average price of tobacco products -0.853 0.355 0.0203

Penetration of TPD+PP products in the market -0.003 0.003 0.3395

AR(1) -0.344 0.225 0.1333

AR(2) 0.226 0.251 0.3727

AR(3) 0.198 0.129 0.1338

MA(1) -0.442 0.215 0.0454

MA(2) -0.557 0.234 0.0216

Table 3.5: Pure time-series model of tobacco consumption for France


% Change in the average price of tobacco products -0.782 0.126 0.0000

Penetration of TPD-PP products in the market -0.001 0.002 0.6589

AR(1) 0.311 0.119 0.0128

AR(12) -0.605 0.110 0.0000

MA(1) -0.972 0.018 0.0000

8 The other element, the “I” in ARIMA, which stands for “Integrated”, in this context means the model is calculated

in first differences (i.e. in changes in values, rather than in levels). 9 Within the broad class of ARIMA models, a specific model is characterised by an order for the autoregressive

components (p), and an order for the moving average component (q). The order simply indicates the lag of the

relationship linking current values to past values, so, for example, an autoregressive term of order two AR(2)

indicates that the current value of a variable depends on the variables’ value observed two periods earlier. 10 For example, the Akaike Information Criterion (AIC) or the Schwarz Bayesian Information Criterion (BIC). For the

purpose of selecting the best ARIMA process we used the BIC statistic, applied iteratively across possible ARIMA

models to an order of up to 3 so as to identify those that perform best. We then inspect the correlogram of the

residuals of the preferred model in order to decide whether the inclusions of additional AR component is

appropriate.


- 12 -

The results presented in Table 3.4 and Table 3.5 indicate that, whilst higher prices are statistically

and significantly associated with lower consumption levels in both France and the UK, the

introduction of TPD2+PP requirements do not appear to have any statistical association with

the number of sticks sold.

3.4.2 Simultaneous equation model

A simultaneous equation model is made up of two different equations that are estimated simultaneously. The

first is a consumption equation similar to that presented in Section 2.4, where changes in the number of sticks

sold are explained by seasonal monthly dummies, percentage changes in price, penetration of TPD2+PP products

in the market and an appropriate time series process. The second equation is a price equation where changes

in prices are modelled by a time-series process and the penetration of TPD2+PP products. Therefore, in a

simultaneous equation model, the potential impact of TPD2+PP on consumption can be broken down into

two different components:

A direct impact of TPD2+PP on consumption (through the consumption equation)

An indirect impact of TPD2+PP on consumptions that feeds through the price channel, i.e. TPD2+PP has

an impact on prices (through the price equation), and prices in turns affect consumptions (through the

consumption equation).

Since a moving average (MA) cannot be estimated within a simultaneous equation framework, the selection

of an appropriate ARMA process for the two equations has been restricted to include only autoregressive

components.

A stylised representation of our modelling approach is presented in the figure below.

Figure 3.3: Simultaneous equations approach to time series estimation of impacts on consumption

The coefficients 𝛼, 𝛽 and 𝛾 are, respectively, the coefficient for the direct impact of TPD2+PP on

consumption, the coefficient for the impact of TPD2+PP on prices, and the coefficient for the impact of prices

on consumption. The following table, sets out the relevant conditions and calculation steps followed in order

to calculate the aggregate TPD2+PP effect and to identify its constituent elements (i.e. direct and indirect,

indirect only, direct only).

Table 3.6: Identifying and calculating the aggregate TPD2/PP penetration effect

Coefficient condition

Identification of aggregate

effect of TPD2+PP

penetration

Calculation of aggregate

effect of TPD2+PP

penetration


- 13 -

𝜶, 𝜷, and 𝜸 are all

statistically significant Both direct and indirect effect 𝜶 + 𝜷 x 𝜸

𝜷 and 𝜸 are statistically

significant Only indirect effect 𝜷 x 𝜸

𝜶 𝒊𝒔 statistically significant Only direct effect 𝜶

Source: Europe Economics

The results of the simultaneous equation model are reported below.

Table 3.7: Results of simultaneous equations time series model of impact of TPD2/PP on tobacco

consumption in the UK and France

Direct impact of

TPD2+PP (𝜶)

Indirect impact

of TPD2+PP (𝜷)

% Price change

impact (𝜸)

Overall impact of

TPD2/PP

UK Statistically

insignificant

Statistically

insignificant

Statistically

insignificant No impact

FR Statistically

insignificant

Statistically

insignificant -0.06** No impact

Note: * = “Significant at 90% confidence level”; ** = “Significant at 95% confidence level”; *** = “Significant at 99% confidence level”.

As we can see from Error! Reference source not found. the model estimates suggest that there is

no statistically significant impact of TPD2 and PP on tobacco consumption either directly or

indirectly via an impact on prices. The only statistically significant relationship is the negative association

between price of tobacco products and the smoking consumption in France (i.e. 𝛾 is statistically significant).

But since 𝛽 is not statistically significant (i.e. is not statistically distinguishable from zero), that means 𝛽 x 𝛾 is

not statistically distinguishable from zero.

Time-series smoking prevalence model for the UK

- 14 -

4 Time-series smoking prevalence

model for the UK

4.1 Introduction

We have publically available data with sufficient frequency to model prevalence impacts for the UK, but not

for France. The data on smoking prevalence for England is reported on a monthly basis by Smoking Toolkit

Study (STS).11 We use such data as a proxy for the smoking prevalence in the UK as a whole. In this section

we first provide an overview of monthly smoking prevalence in England, and we then then present the results

of our statistical analysis. Like for the analysis on tobacco consumption, we first conduct a simple trend

analysis and we then employing more sophisticated models.

4.2 Smoking prevalence England

As we can see from Figure 4.1of smoking prevalence in England has decreases steadily since 2012. Moreover,

the rate of decline in prevalence rate appears to follow a linear trend.

Figure 4.1: Smoking prevalence in England

4.3 Simple trend model

A simple trend model confirms the visual patterns observed in Figure 4.1, i.e. the evolution of smoking

prevalence can be described – in statistically significant terms – as following a declining linear trend. We have

11 See http://www.smokinginengland.info/latest-statistics/

15%

16%

17%

18%

19%

20%

21%

22%

2012M

01

2012M

03

2012M

05

2012M

07

2012M

09

2012M

11

2013M

01

2013M

03

2013M

05

2013M

07

2013M

09

2013M

11

2014M

01

2014M

03

2014M

05

2014M

07

2014M

09

2014M

11

2015M

01

2015M

03

2015M

05

2015M

07

2015M

09

2015M

11

2016M

01

2016M

03

2016M

05

2016M

07

2016M

09

2016M

11

2017M

01

2017M

03

2017M

05

2017M

07

http://www.smokinginengland.info/latest-statistics/


- 15 -

then conducted number of tests to determine whether the series has a break (in levels, in trend, or both) in

May 2016, but none of the tests suggests the presence of a statistically significant break.

4.4 More sophisticated tests

We present below the results of the pure-time series analysis and those of a simultaneous equation model

4.4.1 Time-series models

In our time-series analysis we explain model monthly changes in prevalence with monthly percentage changes

in prices and the penetration of TPD2+PP products in the market. Seasonal patterns in smoking prevalence are

much less marked than those observed for smoking consumption. As a result, the time-series model our

algorithm indicates should be preferred does not include monthly dummies.12

The results of the pure-time series analysis are reported below and indicate that there is no statistically

significant association between prevalence and the introduction of TPD2 and PP requirements.

Table 4.1: Pure time-series model of smoking prevalence in the UK


% Change in the average price of tobacco products 0.244 0.297 0.4157

Penetration of TPD-PP products in the market 0.311 0.209 0.1439

C -0.155 0.097 0.1183

AR(1) -0.800 0.119 0.0000

AR(2) -0.608 0.168 0.0007

AR(3) -0.340 0.109 0.0030

AR(12) 0.2411 0.113 0.0386

4.4.2 Simultaneous equation model

The simultaneous equation model we present here is similar to that presented in Section 3.4.2, and attempts

to identify two separate impacts:

A direct impact of TPD2+PP on smoking prevalence;

An indirect impact of TPD2+PP on smoking prevalence.

The results are reported below and suggest that there is no statistically significant association

(either directly or indirectly) between of TPD2+PP requirements and smoking prevalence in

the UK.

12 We have also run models with monthly dummy variables but such models underperformed (in terms of BIC statistics

output) the model without seasonal dummies that we present here. We note that our algorithm also considered the

possibility that vaping might affect prevalence, but this was never statistically significant in a preferred model.


- 16 -

Table 4.2: Results of simultaneous equations time series model of impact of TPD-PP on smoking

prevalence in the UK

Direct impact of

TPD2+PP

Indirect impact

of TPD2+PP

% Price change

impact

Overall impact of

TPD2/PP

UK Statistically

insignificant

Statistically

insignificant

Statistically

insignificant No impact

Note: * = “Significant at 90% confidence level”; ** = “Significant at 95% confidence level”; *** = “Significant at 99% confidence level”.

Conclusion

- 17 -

5 Conclusion

In this report we have considered the impacts of TPD2 and plain packs requirements, introduced in the UK

and France between 2016 and 2017, upon tobacco consumption and prevalence, in three types of models:

simple linear trend models, time series models and simultaneous equations models. We have found that there

has as yet been no statistically significant impact in the sorts of models we have used:

No statistically significant impacts on prevalence in the UK.

No statistically significant impacts on consumption in the UK or in France.

These results remain provisional as the analysis is limited in time and it should be noted that they are for the

combined impact of TPD2 and plain packs rather than their individual impacts. It might be possible to attempt

to disentangle the impacts of plain packs and TPD2 by using models that deployed other countries that

introduced TPD2 but not plain pack requirements as controls, to explore the possibility that there was a

statistically significant impact in one direction from TPD2 (e.g. a reduction in consumption) but an offsetting

statistically significant impact in the opposite direction from plain packs requirements.

It is also possible that with additional time, the dynamic in the market might change and become more

pronounced. We note, however, that at the time of the impact assessment accompanying TPD2 it was

anticipated that there would be an impact by this point. In the government’s TPD2 impact assessment of

2015 it provides estimates of projected prevalence levels and projected reductions associated with TPD2.

These were for a reduction of 1.9 per cent in both smoking prevalence and smoking consumption over 5

years, applied linearly implying a reduction of 0.38 per cent in the 1st year, equivalent to 0.08 percentage

points reduction in prevalence in the first year (2016) if prevalence would have been 19.7 per cent.13

No such impact is yet identifiable in the data.

13 See paragraphs 73-74 of Tobacco Products Directive (TPD) IA No: 3131, 29/06/2015, https://www.bma.org.uk/-

/media/files/pdfs/working%20for%20change/policy%20and%20lobbying/uk%20consultations/po-tobacco-products-

directive-impact-assesment-2015-09-01.pdf

Conclusion

- 18 -

Appendix

- 19 -

6 Appendix

This section provides a more formal details on the statistical models employed. We also provide the results

for a number of alternative models we have used as a cross check.

6.1 Simple trend models for tobacco consumption (FR and UK) and smoking

prevalence (UK).

6.1.1 Consumption

The estimation results of a simple trend model without break for tobacco consumption in France and the

UK are provided respectively in Table 6.1 and Table 6.2. The result indicate that tobacco consumption in

France does not appear to follow a trend, whilst in the UK is follows a negative trend.

Table 6.1: Trend model of tobacco consumption for France

Dependent Variable: Number of sticks (FR)

Method: Least Squares

Sample: 2012M01 2017M08

Included observations: 68

Variable Coefficient Std. Error t-Statistic Prob.

C 5.20E+09 3.41E+08 15.24557 0.0000

Trend -1982304. 4117631. -0.481418 0.6318

R-squared 0.003499 Mean dependent var 5.04E+09

Adjusted R-squared -0.011599 S.D. dependent var 6.63E+08

Table 6.2: Trend model of tobacco consumption for the UK

Dependent Variable: Number of sticks (UK)


Sample: 2012M01 2017M08



C 5.35E+09 1.12E+08 47.80068 0.0000

Trend -15204327 1349832. -11.26387 0.0000


Adjusted R-squared 0.652624 S.D. dependent var 3.71E+08

Appendix

- 20 -

We report below the results of the trend models in which three types of breaks are introduced (namely a

break in the level, a break in the trend, and a break in the level and the trend). The only break tests to be

statistically significant (at the 99 or 95 per cent confidence level) are the break and in the level and the break

in the trend for the UK.

Table 6.3: Trend model of tobacco consumption for the UK (break in level)



Sample: 2012M01 2017M08



C 5.60E+09 1.41E+08 39.82594 0.0000

Trend -19045013 1897574. -10.03650 0.0000

Dummy 2.42E+08 87804627 2.754543 0.0076



Table 6.4: Trend model of tobacco consumption for the UK (break in trend)



Sample: 2012M01 2017M08



C 5.60E+09 1.41E+08 39.62268 0.0000

Trend -19121538 1910859. -10.00677 0.0000

Dummy*Trend 2298611. 829213.5 2.772037 0.0073



Table 6.5: Trend model of tobacco consumption for the UK (break in level and trend)



Sample: 2012M01 2017M08



C 5.61E+09 1.44E+08 39.06184 0.0000

Trend -19147176 1938408. -9.877784 0.0000

Dummy -1.40E+08 1.22E+09 -0.114939 0.9089

Dummy*Trend 3621667. 11541262 0.313802 0.7547



Table 6.6: Trend model of tobacco consumption for France (break in level)



Sample: 2012M01 2017M08


Appendix

- 21 -


C 5.35E+09 4.52E+08 11.83771 0.0000

Trend -4295360. 6104561. -0.703631 0.4842

Dummy 1.46E+08 2.82E+08 0.515670 0.6078



Table 6.7: Trend model of tobacco consumption for France (break in trend)



Sample: 2012M01 2017M08



C 5.35E+09 4.55E+08 11.75721 0.0000

Trend -4291313. 6151783. -0.697572 0.4879

Dummy*Trend 1354922. 2669554. 0.507546 0.6135



Table 6.8: Trend model of tobacco consumption for France (break in level and trend)



Sample: 2012M01 2017M08



C 5.35E+09 4.62E+08 11.57192 0.0000

Trend -4198566. 6240303. -0.672814 0.5035

Dummy 5.08E+08 3.93E+09 0.129159 0.8976

Dummy*Trend -3431366. 37154703 -0.092353 0.9267



6.1.2 Prevalence

Finally we provide here the same trend model and break test for UK smoking prevalence. The models show

that smoking prevalence can be described as following declining trend without any statistically significant

breaks.

Table 6.9: Trend model of smoking prevalence (UK)

Dependent Variable: Smoking prevalence (UK)


Sample: 2012M01 2017M08



C 22.01904 0.608403 36.19151 0.0000

Trend -0.040297 0.007343 -5.488098 0.0000

Appendix

- 22 -

R-squared 0.313353 Mean dependent var 18.77510

Adjusted R-squared 0.302949 S.D. dependent var 1.423472



Date: 11/21/17 Time: 14:02

Sample: 2012M01 2017M08



C 21.75079 0.806471 26.97035 0.0000

Trend -0.036213 0.010886 -3.326514 0.0014

Dummy -0.257186 0.503731 -0.510563 0.6114





Date: 11/21/17 Time: 14:02

Sample: 2012M01 2017M08



C 21.71900 0.811594 26.76092 0.0000

Trend -0.035737 0.010965 -3.259176 0.0018

Dummy*Trend -0.002676 0.004758 -0.562359 0.5758





Date: 11/21/17 Time: 14:02

Sample: 2012M01 2017M08



C 21.65254 0.820577 26.38698 0.0000

Trend -0.034858 0.011083 -3.145115 0.0025

Dummy 4.812869 6.982961 0.689230 0.4932

Dummy*Trend -0.048039 0.065990 -0.727971 0.4693



Appendix

- 23 -

6.2 Pure time-series models for tobacco consumption (FR and UK) and

smoking prevalence (UK).

In Sections 6.2.1 and 6.2.2 we present the detailed results of our preferred time-series models for tobacco

consumption and smoking prevalence. In Section 6.2.3 we present the result we obtain with an alternative

modelling approach.

6.2.1 Benchmark models for Consumption

Table 6.10: Pure time series-model of tobacco consumption (UK)

Dependent Variable: %Change in number of sticks (UK)


Sample (adjusted): 2012M05 2017M08

Included observations: 64 after adjustments

Convergence achieved after 22 iterations

HAC standard errors & covariance


Jan dummy -11.88168 1.134776 -10.47051 0.0000

Feb dummy -1.882840 0.887128 -2.122399 0.0393

Mar dummy 10.94947 1.046088 10.46707 0.0000

Apr dummy -6.482104 0.696819 -9.302415 0.0000

May dummy 11.33563 0.817969 13.85825 0.0000

Jun dummy -4.081629 0.481215 -8.481925 0.0000

Jul dummy 3.077330 1.002184 3.070625 0.0036

Aug dummy 0.865219 0.770582 1.122812 0.2675

Sep dummy -4.525822 0.775342 -5.837197 0.0000

Oct dummy -5.020536 0.554933 -9.047110 0.0000

Nov dummy 1.878206 0.368610 5.095371 0.0000

Dec dummy 7.385995 1.139673 6.480801 0.0000

% change in price -0.853113 0.353680 -2.412103 0.0200

TPD2+PP penetration -0.002638 0.002730 -0.966380 0.3390

AR(1) -0.343725 0.224639 -1.530123 0.1330

AR(2) 0.225724 0.250418 0.901386 0.3722

AR(3) 0.197550 0.129326 1.527540 0.1336

MA(1) -0.443491 0.214633 -2.066277 0.0446

MA(2) -0.555654 0.233774 -2.376890 0.0218



S.E. of regression 1.399288 Akaike info criterion 3.751334

Sum squared resid 88.11031 Schwarz criterion 4.392252

Log likelihood -101.0427 Hannan-Quinn criter. 4.003824

Durbin-Watson stat 1.962075 Wald F-statistic 172.2708

Prob(Wald F-statistic) 0.000000

Table 6.11: Pure time series-model of tobacco consumption (FR)

Dependent Variable: %Change in number of sticks (FR)


Date: 11/17/17 Time: 17:56






Appendix

- 24 -

Jan dummy -0.230955 0.006354 -36.34520 0.0000

Feb dummy 0.033976 0.004788 7.095887 0.0000

Mar dummy 0.266615 0.003368 79.16514 0.0000

Apr dummy -0.188187 0.005121 -36.75041 0.0000

May dummy 0.023034 0.003274 7.035746 0.0000

Jun dummy 0.311199 0.008507 36.58154 0.0000

Jul dummy -0.245576 0.005825 -42.16204 0.0000

Aug dummy -0.057018 0.005956 -9.573870 0.0000

Sep dummy 0.325113 0.007687 42.29196 0.0000

Oct dummy -0.227999 0.005099 -44.71821 0.0000

Nov dummy -0.020638 0.004501 -4.585158 0.0000

Dec dummy 0.282196 0.007989 35.32237 0.0000

% Change in price -0.782411 0.125536 -6.232572 0.0000


AR(12) -0.605147 0.109540 -5.524453 0.0000

AR(1) 0.311365 0.119155 2.613108 0.0128

MA(1) -0.972843 0.018013 -54.00813 0.0000



S.E. of regression 0.015809 Akaike info criterion -5.207978

Sum squared resid 0.009498 Schwarz criterion -4.587530

Log likelihood 160.2194 Hannan-Quinn criter. -4.968046



6.2.2 Benchmark models for Prevalence

Table 6.12: Pure time series-model of smoking prevalence (UK)

Dependent Variable: Changes in prevalence (UK)







% Change in price 0.243294 0.293681 0.828429 0.4114


C -0.088166 0.074690 -1.180436 0.2434

AR(12) 0.282771 0.125278 2.257157 0.0284

MA(1) -0.999582 0.089587 -11.15769 0.0000

R-squared 0.596263 Mean dependent var -0.080443





F-statistic 18.46073 Durbin-Watson stat 2.189984

Prob(F-statistic) 0.000000 Wald F-statistic 4.566011


Appendix

- 25 -

6.2.3 Alternative modelling approaches.

The time-series modelling approach used to produce the results set out in Section 6.2.1 6.2.2 relies on

expressing the dependent variable (tobacco consumption and prevalence) in terms of percentage change.

This transformation ensures that dependent variables are stationary and can therefore be modelled

meaningfully. We then control for seasonal patterns in the data with the use of dummy variables. It is

however important to stress that models in which the dependent variable are expressed in differences (or,

in our case, in percentage differences) are less likely to find a statistically significant relationship between

variables than models expressed in levels. In other words, if we find a statistically significant relationship in a

model in difference we can be quite certain that such a relationship exists. However, if we fail to find such

relationship to be statistically significant in differences, we might still be able to find it to be significant in a

model expressed in levels (even though the risk of the statistical relationship being sporous is higher).

As a cross check we have therefore analysed a number of alternative models where the dependent variables

are expressed in levels as opposed to percentage changes. Such modelling approach would still require some

transformation of the dependent variable to ensure stationarity. In many cases we find that the dependent

variables are trend-stationary, i.e. the variables themselves are non-stationary because they follow a linear

trend, but their fluctuations around the trend are stationary. Furthermore, since deviations from the trend

have a seasonal component, we can de-seasonalise them with the use dummy variables. Therefore after being

de-trended and de-seasonalised, the variables can be analysed in a meaningful way.

One advantage of this approach is that it reduce the risk of what is called “over-differencing”. Most time

series models (and in particular the form of time series models we use here) work best when they have been

rendered “stationary”, i.e. normalised such that the mean, variance, autocorrelation, etc. of the normalised

series are all constant over time. In a number of time series setting, differencing (i.e. considering changes in

rather than the levels of variables) is a useful and common technique to render an otherwise “non-stationary”

series stationary. However, there is some risk that in the process of differencing we erase the correlation

between variables that the time series model is seeking to identify (i.e. we “over-difference”). The alternative

approach we set out here, where instead of differencing we use levels adjusted for trends and seasonal

patterns, checks for the sorts of correlations that over-differencing might have erased.

A disadvantage of this approach — and the key reason we regard the approach we set out in the main body

of this report as preferred — is that this approach is more complex and requires more judgement to execute

(e.g. in determining which approach to take to detrending or seasonalising). Furthermore, our approach in

the main body is slightly more demanding of the data than the alternative set out here, and thus is appropriate

in testing the robustness of a perhaps-counterintuitive or at least unintended result such as that in our simple

trend model — namely that TPD2+PP is associated with higher consumption. If, on the other hand, our

simple trend model had identified that TPD2+PP was associated with a reduction in consumption, there might

have been more of a case for considering our alternative approach as an intermediate step — i.e. we would

shift from simple model to normalised levels model to differenced model and see how robust that correlation

was to more and more demanding tests.

To being with, we report below the result of the unit root-test for the following (unadjusted) variables:

Number of sticks in the UK.

Prevalence in the UK.

Number of sticks in France.

The tests confirms that tobacco consumption and smoking prevalence in the UK are non-stationary, but they

are trend-stationary. However, tobacco consumption in France appears to be stationary (at the 95 percent

confidence level) even without de-trending.

Appendix

- 26 -

Table 6.13: ADF test for tobacco consumption in the UK

t-Statistic Prob.*

Augmented Dickey-Fuller test statistic (Constant) -1.797211 0.3789

Augmented Dickey-Fuller test statistic (Constant, Linear Trend) -7.933299 0.0000

Table 6.14: Table 6.15: ADF test for smoking prevalence in the UK

t-Statistic Prob.*



Table 6.16: ADF test for tobacco consumption in France

t-Statistic Prob.*



Based on the unit-root tests conducted above we have carried out the following variable transformations:

We have de-trended and de-seasonalised the number of sticks in the UK and smoking prevalence in the

UK.

We have de-seasonalised the number of sticks in France.

The dependent variables after such adjustments are reported in the figures below.

Figure 6.1: Number of sticks (France and UK)

-400,000,000

-300,000,000

-200,000,000

-100,000,000

-

100,000,000

200,000,000

300,000,000

400,000,000

2012M

01

2012M

03

2012M

05

2012M

07

2012M

09

2012M

11

2013M

01

2013M

03

2013M

05

2013M

07

2013M

09

2013M

11

2014M

01

2014M

03

2014M

05

2014M

07

2014M

09

2014M

11

2015M

01

2015M

03

2015M

05

2015M

07

2015M

09

2015M

11

2016M

01

2016M

03

2016M

05

2016M

07

2016M

09

2016M

11

2017M

01

2017M

03

2017M

05

2017M

07

Number of sticks in the UK (de-trended and seasonally adjusted)

Number of sticks in the FR (seasonally adjusted)

Appendix

- 27 -

Figure 6.2: Smoking prevalence (UK)

We have then modelled the above adjusted variables with the following explanatory variables: percentage

changes in the average price of cigarettes, penetration of TPD2+PP products, and an optimally selected ARMA

process. The results of such models are reported below.

Table 6.17: Alternative time-series model of tobacco consumption (UK)

Dependent Variable: Number of sticks (UK) – de-trended and de-seasonalised







% Change in price -1.20E+09 1.36E+09 -0.881870 0.3823

TPD2+PP penetration 1.05E+08 33879280 3.108016 0.0032

C -24196093 13080108 -1.849839 0.0706

AR(1) 0.881033 0.072432 12.16351 0.0000

AR(2) -0.584877 0.109486 -5.342017 0.0000

AR(12) -0.292297 0.075784 -3.856965 0.0003

MA(1) -0.640215 0.054947 -11.65140 0.0000

MA(2) 0.970106 0.029148 33.28233 0.0000

R-squared 0.534687 Mean dependent var -14640942

Adjusted R-squared 0.465385 S.D. dependent var 69626918

S.E. of regression 50909392 Akaike info criterion 38.46272

Sum squared resid 1.22E+17 Schwarz criterion 38.75469





Table 6.18: Alternative time-series model of tobacco consumption (FR)

Dependent Variable: Number of sticks (FR) –de-seasonalised


-3%

-2%

-1%

0%

1%

2%

3%

2012M

01

2012M

03

2012M

05

2012M

07

2012M

09

2012M

11

2013M

01

2013M

03

2013M

05

2013M

07

2013M

09

2013M

11

2014M

01

2014M

03

2014M

05

2014M

07

2014M

09

2014M

11

2015M

01

2015M

03

2015M

05

2015M

07

2015M

09

2015M

11

2016M

01

2016M

03

2016M

05

2016M

07

2016M

09

2016M

11

2017M

01

2017M

03

2017M

05

2017M

07

Prevalence (de-trended and seasonally adjusted)

Appendix

- 28 -






% Change in price 1.48E+09 1.05E+09 1.414638 0.1634

TPD2+PP penetration 35391179 43936200 0.805513 0.4243

C -52446992 22839261 -2.296352 0.0259

AR(1) 0.672015 0.085446 7.864783 0.0000

AR(12) -0.145444 0.059326 -2.451583 0.0178









Table 6.19: Alternative time-series model of smoking prevalence (UK)

Dependent Variable: Smoking prevalence (UK) – de-trended and de-seasonalised




HAC standard errors & covariance (Bartlett kernel, Newey-West fixed

bandwidth = 4.0000)

Variable Coefficient Std. Error t-Statistic Prob. % Change in price -7.770281 20.33475 -0.382118 0.7036


C 0.035668 0.155353 0.229592 0.8191 R-squared 0.003427 Mean dependent var -0.002007

Adjusted R-squared -0.027716 S.D. dependent var 0.994814







We see from Table 6.17, that under this alternative modelling approach the introduction of TPD2+PP in the

UK is statistically associated (at the 95 per cent confidence level) with an increase in the number of sticks

sold. In all other models the introduction of TPD2+PP remains statistically insignificant. We also note, in

Table 6.19 that, after de-trending and de-seasonalising, there no time series structure explaining the evolution

of smoking prevalent in the UK.

Appendix

- 29 -

6.2.4

6.3 Simultaneous equation models of tobacco consumption (FR and UK) and

smoking prevalence (UK).

In Sections 6.3.1 and 6.3.2 we present the detailed results of our simultaneous-equation approach. In

Section 6.3.3 we present the result we obtain when we model the consumption equation and the

prevalence with the alternative approach described in 6.2.3, ie by using de-trended and de-seasonalised

series.

6.3.1 Consumption

Below we first present estimation results for the two separate equations (consumption equation and the

price equation) constituting the system model for the UK. We then provide the estimation output of the

system as a whole (the coefficients of the systems are coded from C(1) to C(21)).

Table 6.20: Consumption equation (UK)

Dependent Variable: %Change in number of sticks (UK)







Jan dummy – C(1) -0.129707 0.012847 -10.09644 0.0000

Feb dummy – C(2) -0.014292 0.009551 -1.496381 0.1412

Mar dummy – C(3) 0.102621 0.011478 8.940368 0.0000

Apr dummy – C(4) -0.067798 0.006445 -10.51873 0.0000

May dummy – C(5) 0.108252 0.008165 13.25799 0.0000

Jun dummy – C(6) -0.040937 0.005140 -7.964950 0.0000

Jul dummy – C(7) 0.023875 0.010011 2.384891 0.0212

Aug dummy – C(8) 0.007914 0.006808 1.162371 0.2510

Sep dummy – C(9) -0.047890 0.007511 -6.376170 0.0000

Oct dummy – C(10) -0.049283 0.004172 -11.81409 0.0000

Nov dummy – C(11) 0.015665 0.003060 5.118889 0.0000

Dec dummy – C(12) 0.076023 0.011796 6.444657 0.0000

% Change in price – C(13) -0.023253 0.636968 -0.036506 0.9710

TPD2+PP penetration – C(14) -0.000703 0.003496 -0.200984 0.8416

AR(1) – C(15) -0.505837 0.112412 -4.499826 0.0000

AR(2) – C(16) -0.126897 0.155517 -0.815973 0.4186

AR(3) – C(17) 0.061557 0.099467 0.618869 0.5390








Inverted AR Roots .22 -.36-.39i -.36+.39i

Appendix

- 30 -

Table 6.21: Price equation (UK)

Dependent Variable: %Change in price (UK)








C – C(19) 0.000993 0.001441 0.689420 0.4937

AR(1) – C(20) -0.192167 0.087678 -2.191746 0.0330

AR(12) – C(21) 0.700237 0.074611 9.385222 0.0000









Inverted AR Roots .96 .83-.48i .83+.48i .47-.84i

.47+.84i -.02-.97i -.02+.97i -.50+.84i

-.50-.84i -.86+.48i -.86-.48i -.99

Table 6.22: Simultaneous equation model of consumption (UK)

System (UK)

Estimation Method: Iterative Least Squares

Sample: 2012M05 2017M08


Total system (unbalanced) observations 119


Coefficient Std. Error t-Statistic Prob.

C(1) -0.129707 0.011546 -11.23370 0.0000

C(2) -0.014292 0.008334 -1.715036 0.0895

C(3) 0.102621 0.009285 11.05201 0.0000

C(4) -0.067798 0.008034 -8.439080 0.0000

C(5) 0.108252 0.007947 13.62189 0.0000

C(6) -0.040937 0.007101 -5.764998 0.0000

C(7) 0.023875 0.009725 2.454987 0.0159

C(8) 0.007914 0.006993 1.131649 0.2605

C(9) -0.047890 0.007552 -6.341432 0.0000

C(10) -0.049283 0.007546 -6.531261 0.0000

C(11) 0.015665 0.007866 1.991624 0.0492

C(12) 0.076023 0.007732 9.831942 0.0000

C(13) -0.023261 0.655719 -0.035474 0.9718

C(14) -0.000703 0.005086 -0.138161 0.8904

C(15) -0.505837 0.141811 -3.566974 0.0006

C(16) -0.126898 0.165013 -0.769016 0.4437

C(17) 0.061557 0.132164 0.465759 0.6424

C(18) -0.000646 0.001188 -0.544292 0.5875

C(19) 0.000993 0.001127 0.881118 0.3804

C(20) -0.192167 0.080750 -2.379787 0.0193

Appendix

- 31 -

C(21) 0.700237 0.078014 8.975804 0.0000

For France, estimation results for the two separate system equations and for the system as a whole are

reported below (the coefficients of the systems are coded from C(1) to C(18)).

Table 6.23: Consumption equation (FR)

Dependent Variable: %Change in number of sticks (FR)







Jan dummy – C(1) -0.227381 0.008510 -26.71946 0.0000

Feb dummy – C(2) 0.032137 0.006768 4.748507 0.0000

Mar dummy – C(3) 0.267953 0.009330 28.72091 0.0000

Apr dummy – C(4) -0.192211 0.007576 -25.37172 0.0000

May dummy – C(5) 0.025326 0.006909 3.665796 0.0006

Jun dummy – C(6) 0.307333 0.010322 29.77432 0.0000

Jul dummy – C(7) -0.243539 0.006484 -37.55881 0.0000

Aug dummy – C(8) -0.056444 0.009873 -5.716893 0.0000

Sep dummy – C(9) 0.326064 0.017642 18.48181 0.0000

Oct dummy – C(10) -0.233604 0.010837 -21.55628 0.0000

Nov dummy – C(11) -0.017034 0.009446 -1.803236 0.0773

Dec dummy – C(12) 0.275462 0.008504 32.39371 0.0000

% Change in price – C(13) -0.600980 0.421292 -1.426517 0.1598

TPD2+PP penetration – C(14) 0.004020 0.002946 1.364588 0.1784

AR(1) – C(15) -0.435354 0.117535 -3.704021 0.0005








Inverted AR Roots -.44

Table 6.24: Price equation (FR)

Dependent Variable: %Change in price (FR)





HAC standard errors & covariance Variable Coefficient Std. Error t-Statistic Prob.


C – C(17) 0.000403 0.000985 0.409066 0.6841

AR(1) – C(18) -0.210309 0.089010 -2.362770 0.0218

AR(9) – C(19) 0.309955 0.173792 1.783482 0.0801


Appendix

- 32 -








Table 6.25: Simultaneous equation model of consumption (FR)

System (FR)


Sample: 2012M03 2017M08





C(1) -0.227381 0.010381 -21.90387 0.0000

C(2) 0.032137 0.010325 3.112666 0.0024

C(3) 0.267953 0.009492 28.23010 0.0000

C(4) -0.192211 0.009364 -20.52683 0.0000

C(5) 0.025326 0.009329 2.714751 0.0078

C(6) 0.307333 0.009467 32.46353 0.0000

C(7) -0.243539 0.009329 -26.10667 0.0000

C(8) -0.056444 0.009349 -6.037307 0.0000

C(9) 0.326064 0.010476 31.12552 0.0000

C(10) -0.233604 0.010204 -22.89266 0.0000

C(11) -0.017034 0.010191 -1.671507 0.0976

C(12) 0.275462 0.010530 26.16012 0.0000

C(13) -0.600979 0.242801 -2.475194 0.0149

C(14) 0.004020 0.005583 0.720097 0.4731

C(15) -0.435355 0.132462 -3.286645 0.0014

C(16) 0.003093 0.002497 1.238641 0.2182

C(17) 0.000403 0.001246 0.323588 0.7469

C(18) -0.210310 0.119637 -1.757890 0.0817

C(19) 0.309955 0.089476 3.464104 0.0008

6.3.2 Prevalence

The estimation results of the two equations constituting the prevalence system model for the UK are

reported below together with the estimation output of the model as a whole.

Table 6.26: Prevalence equation (UK)

Dependent Variable: Change in prevalence (UK)







% Change in price – C(1) 24.39122 29.71090 0.820952 0.4157


C – C(3) -0.155055 0.097487 -1.590525 0.1183

Appendix

- 33 -

AR(1) – C(4) -0.800223 0.119349 -6.704904 0.0000

AR(2) – C(5) -0.608348 0.168030 -3.620484 0.0007

AR(3) – C(6) -0.340353 0.108971 -3.123342 0.0030

AR(12) – C(7) 0.241178 0.113386 2.127054 0.0386

R-squared 0.490603 Mean dependent var -0.080443








Inverted AR Roots .80 .68+.45i .68-.45i .35-.79i

.35+.79i -.06+.94i -.06-.94i -.47-.79i

-.47+.79i -.81-.43i -.81+.43i -.95

Table 6.27: Price equation (UK)

Dependent Variable: %Change in price (UK)








C – C(9) 0.000993 0.001441 0.689420 0.4937

AR(1) – C(10) -0.192167 0.087678 -2.191746 0.0330

AR(12) – C(11) 0.700237 0.074611 9.385222 0.0000









Inverted AR Roots .96 .83-.48i .83+.48i .47-.84i

.47+.84i -.02-.97i -.02+.97i -.50+.84i

-.50-.84i -.86+.48i -.86-.48i -.99

Table 6.28: Simultaneous equation model of prevalence (UK)

System (UK)


Sample: 2013M02 2017M08


Total system (balanced) observations 110



C(1) 24.39086 39.95897 0.610398 0.5430

Appendix

- 34 -

C(2) 0.311120 0.270258 1.151195 0.2524

C(3) -0.155054 0.135334 -1.145714 0.2547

C(4) -0.800223 0.137375 -5.825118 0.0000

C(5) -0.608349 0.163921 -3.711226 0.0003

C(6) -0.340354 0.137449 -2.476216 0.0150

C(7) 0.241178 0.110446 2.183684 0.0313

C(8) -0.000646 0.001188 -0.544292 0.5875

C(9) 0.000993 0.001127 0.881118 0.3804

C(10) -0.192167 0.080750 -2.379787 0.0192

C(11) 0.700237 0.078014 8.975804 0.0000

6.3.3 Simultaneous equation under alternative approach.

The consumption equations for France and the UK under the alternative approach are reported below

(notice that, for both countries, the adjusted consumption variables are best explained by the same AR

structure, ie one with one AR(1) and one AR(12) term).

Table 6.29: Alternative consumption equation (UK)

Dependent Variable: Number of sticks (de-trended and de-seasonalised)







% Change in price – C(1) -1.20E+09 1.29E+09 -0.933094 0.3553

TPD2+PP penetration – C(2) 1.05E+08 29292335 3.594679 0.0007

C – C(3) -24195560 14706038 -1.645281 0.1062

AR(1) – C(4) 0.465818 0.145384 3.204064 0.0024

AR(12) – C(5) -0.149243 0.088668 -1.683170 0.0986


Adjusted R-squared 0.359560 S.D. dependent var 69626918







Table 6.30: Alternative consumption equation (FR)

Dependent Variable: Number of sticks (de-seasonalised)





HAC standard errors & covariance Variable Coefficient Std. Error t-Statistic Prob. % Change in price – C(1) 1.48E+09 1.05E+09 1.414638 0.1634

TPD2+PP penetration – C(2) 35391179 43936200 0.805513 0.4243 C – C(3) -52446992 22839261 -2.296352 0.0259

Appendix

- 35 -

AR(1) – C(4) 0.672015 0.085446 7.864783 0.0000 AR(12) – C(5) -0.145444 0.059326 -2.451583 0.0178









The price equations for France and the UK are the same as in Table 6.21 and Table 6.24. However to

facilitate the interpretation of the simultaneous equations models the coefficients of the price equations can

be renumbered as follows:

Dependent variables and coefficients codes of price equation for the UK

TPD2+PP penetration – C(6)

C – C(7)

AR(1) – C(8)

AR(12) – C(9)

Dependent variables and coefficients codes of price equation for France


C – C(7)

AR(1) – C(8)

AR(9) – C(9)

The estimation output of the simultaneous equating models for consumptions are below.

Table 6.31: Alternative simultaneous equation model of consumption (UK)

System


Sample: 2013M02 2017M08


Total system (balanced) observations 110



C(1) -3.22E+08 8.78E+08 -0.366962 0.7144

C(2) 92892040 46207404 2.010328 0.0471

C(3) -24532019 12464725 -1.968115 0.0518

C(4) 0.490354 0.124495 3.938731 0.0002

C(5) -0.152944 0.100923 -1.515453 0.1328

C(6) -0.000646 0.001188 -0.544296 0.5874

C(7) 0.000993 0.001127 0.881163 0.3803

C(8) -0.192167 0.080750 -2.379786 0.0192

C(9) 0.700237 0.078014 8.975804 0.0000

Table 6.32: Alternative simultaneous equation model of consumption (FR)

System:


Sample: 2012M11 2017M08




Appendix

- 36 -


C(1) 2.25E+09 8.89E+08 2.528743 0.0129

C(2) 15596221 73895746 0.211057 0.8333

C(3) -41315001 25360103 -1.629134 0.1063

C(4) 0.679995 0.098163 6.927215 0.0000

C(5) -0.147897 0.071537 -2.067419 0.0412

C(6) 0.003093 0.002497 1.238641 0.2183

C(7) 0.000403 0.001246 0.323588 0.7469

C(8) -0.210310 0.119637 -1.757890 0.0817

C(9) 0.309955 0.089476 3.464104 0.0008

The statistical significance of the coefficient C(2) in Table 6.31 confirms results we obtained with the

alternative time-series model of Table 6.17:, in the UK the introduction of TPD2+PP is associated with an

increase in tobacco consumption. However, the alternative simultaneous equation model for France indicates

that there is is no statistically significant relationship between TPD2+PP and tobacco consumption.

Finally, the prevalence equations for the UK under the alternative approach is the same provided in Table

6.19, whilst the price equation is the same as in Table 6.21. For the same of interpretation of the simultaneous

equation system the coefficients of the two equations (prevalence equation and price equation) are labelled

as follows:

Dependent variables and coefficients codes of the prevalence equation for the UK

% Change in price – C(1)


C – C(3)

Dependent variables and coefficients codes of price equation for the UK


C – C(5)

AR(1) – C(6)

AR(9) – C(7)

As Table 6.33 shows there is no statistically significant relationship between prevalence and the introduction

of TPD2+PP in the UK.

Table 6.33: Alternative simultaneous equation model of prevalence (UK)

System:


Sample: 2012M02 2017M08





C(1) -7.770281 19.85453 -0.391361 0.6963

C(2) -0.141121 0.460649 -0.306353 0.7599

C(3) 0.035668 0.147073 0.242516 0.8088

C(4) -0.000646 0.001188 -0.544292 0.5873

C(5) 0.000993 0.001127 0.881118 0.3801

C(6) -0.192167 0.080750 -2.379787 0.0190

C(7) 0.700237 0.078014 8.975804 0.0000

tpd2 and standardised tobacco packaging what impacts have … · 2020-05-12 · updating health...

Documents